I'll assume the graph, denoted $G$, is connected, and that the surface $Q$ is closed, connected, and oriented; I believe those are the prerequisites for the definition of genus.
It is not true that such a curve exists. For example, in the torus $T=S^1 \times S^1$ consider the circle graph $G = S^1 \times \{\text{1 point}\} \subset T$. The complement $T-G$ is an open annulus, homeomorphic to $S^1 \times \{\text{an open interval}\}$, and every simple closed curve in an open annulus separates the annulus into two components.
I would suggest a different way to prove that theorem about minimal genus embeddings. If some component of $Q-G$ is not homeomorphic to an open disc then there exists a simple closed curve $C \subset Q-G$ such that $C$ is homotopic to a closed curve in $G$ (possibly not simple), but $C$ is not homotopic in $Q-G$ to a point. By doing surgery along $C$ --- cutting $Q$ along $C$ and gluing in two discs --- you get a smaller genus embedding of $G$.